Power business data forecasting methods, devices, terminal equipment and storage media

Abstract

This application relates to the field of data processing technology, and in particular to a method, apparatus, terminal equipment, and storage medium for predicting power business data. The method includes: acquiring electricity consumption status data for a time period to be measured; constructing comprehensive features based on the electricity consumption status data; acquiring historical similar time periods for the time period to be measured based on the comprehensive features; inputting the historical similar time periods into a pre-trained prediction model for parameter adjustment; and outputting predicted power business data for the time period to be measured based on the adjusted parameters. The power business data includes power load data, power price data, and generator output power data. This allows for more accurate prediction of power business data, thereby better serving the adjustment of power dispatching schemes.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, apparatus, terminal equipment and storage medium for predicting power business data. Background Technology

[0002] Currently, power business data forecasting largely considers the impact of historical similar days on business data fluctuations. There are three main types of similar day models: the first is the traditional similar day model based on feature similarity; the second is the similar day model based on intelligent optimization algorithms; and the third is the similar day model based on clustering algorithms. However, the first method, when determining the final set of similar days, usually relies on experience to set a fixed proportion, such as selecting the top 10% or top 50% of samples by similarity, lacking objective statistical theoretical support. This subjective threshold is prone to significant errors. While the second method solves the problem of traditional similar day models relying entirely on human experience for prediction accuracy, it does not consider the similarity of business data peaks and troughs or grid congestion when considering the similarity of various factors, nor does it consider the order relationship within the similarity of various factors. In the third method, existing clustering algorithms ignore the decision-maker's subjective perception, and existing clustering algorithms for selecting similar days usually only complete preliminary classification without further evaluation and selection of samples within the same category. Currently, the traditional method of predicting business data using similar day models often results in low accuracy in predicting power business data due to insufficient consideration of similar day characteristics or excessive subjectivity in the selection of similar days. Consequently, it is impossible to effectively adjust power dispatching schemes. Summary of the Invention

[0003] In view of this, embodiments of this application provide a method for predicting power business data, which can achieve more accurate data prediction and thus effectively adjust power dispatching schemes.

[0004] In a first aspect, embodiments of this application provide a method for predicting power business data, including:

[0005] Obtain power consumption data for the time period to be tested;

[0006] Based on the electricity consumption data, a comprehensive feature is constructed, and based on the comprehensive feature, historical similar time periods of the time period to be measured are obtained;

[0007] The historical similar time period is input into the pre-trained prediction model for parameter adjustment, and then the adjusted prediction model is used to output the predicted power business data for the time period to be tested.

[0008] Based on the predicted power business data, the power dispatching scheme is adjusted. The power business data includes power load data, power price data, and generator output power data.

[0009] In some embodiments, the electricity status data includes meteorological data, load data, power generation data, and power flow restriction data;

[0010] The step of constructing comprehensive features based on the electricity consumption status data includes:

[0011] Calculate the similarity of electricity consumption characteristics based on the meteorological data, the load data, and the power generation data;

[0012] The time similarity is obtained based on the time period to be tested;

[0013] Calculate the blockage similarity based on the aforementioned power flow restriction data;

[0014] The blockage similarity, time similarity, and electricity consumption feature similarity are sorted to obtain the corresponding order relationship. The comprehensive feature is calculated by combining the blockage similarity, time similarity, electricity consumption feature similarity, and the corresponding order relationship.

[0015] In some embodiments, obtaining historical similar time periods of the time period to be tested based on the comprehensive features includes:

[0016] According to a preset time interval, multiple historical time periods are obtained, along with historical electricity consumption status data and historical power business data corresponding to the multiple historical time periods;

[0017] Historical characteristics are obtained based on the historical electricity consumption status data and historical power business data;

[0018] Based on the historical features and the comprehensive features, obtain multiple historical similar time periods with the highest similarity to the time period to be tested.

[0019] In some embodiments, obtaining multiple historically similar time periods with the highest similarity to the time period to be tested, based on the historical features and the comprehensive features, includes:

[0020] Based on the comprehensive features and each of the historical features, the comprehensive similarity of each historical period relative to the time period to be tested is obtained;

[0021] The combined features are clustered using the mean clustering algorithm to obtain multiple sets of historical time periods. The historical time periods in the set of historical time periods with the highest comprehensive similarity are the first similar time periods.

[0022] Multiple first similar time periods are selected through a two-stage screening using an attention-enhanced mutual information-kernel density estimation method to obtain multiple second similar time periods.

[0023] In some embodiments, the mutual information-kernel density estimation method enhanced by the attention mechanism performs a two-stage screening of multiple first similar time periods to obtain multiple second similar time periods, including:

[0024] Based on the first similar time period, obtain the corresponding time, electricity consumption characteristics and congestion time sequence feature vectors;

[0025] Normalize all the aforementioned time-series feature vectors to obtain dimensionless features;

[0026] The weights of each dimension in the dimensionless features are determined by an attention mechanism, and the nonlinear relationship between the first similarity time and the time period to be tested is calculated.

[0027] The mutual information values ​​of all the first similar time periods are calculated by kernel density estimation, and a probability density function of the mutual information values ​​is constructed to identify the mutual information value interval corresponding to the peak of the probability density function.

[0028] The first similar time period within the mutual information value interval that exceeds the half-width of the probability density function is designated as the second similar time period.

[0029] In some embodiments, the prediction model is one of a time series model, a neural network model, a distributed gradient model, and a ensemble machine learning model such as Bagging-XGBoost.

[0030] In some embodiments, the adjustment of the power dispatching scheme based on the predicted power business data includes power load data, power price data, and generator output power data, including:

[0031] Based on the range of the predicted power business data, the power generation, power consumption, power distribution, generator set combination, peak shaving, frequency regulation, and reserve plan arrangements for the time period to be tested are adjusted to adjust the power dispatching scheme.

[0032] Secondly, this application also provides a power business data forecasting device, comprising:

[0033] The data acquisition module is used to acquire electricity consumption status data for the time period to be tested;

[0034] The feature construction module is used to construct comprehensive features based on the electricity consumption status data, and to obtain historical similar time periods for the time period to be measured based on the comprehensive features.

[0035] The model adjustment module is used to input the historical similar time period into the pre-trained prediction model for parameter adjustment, and the prediction model after parameter adjustment outputs the predicted power business data for the time period to be tested.

[0036] The scheduling and adjustment module is used to adjust the power dispatching scheme based on the predicted power business data, which includes power load data, power price data, and generator output power data.

[0037] Thirdly, this application also provides a terminal device, the terminal device including a processor and a memory, the memory storing a computer program, and the processor for executing the computer program to implement the power business data prediction method.

[0038] Fourthly, this application also provides a computer-readable storage medium storing a computer program that, when executed on a processor, implements the power business data prediction method described above.

[0039] The embodiments of this application have the following beneficial effects:

[0040] This embodiment acquires electricity consumption status data for a test time period; constructs comprehensive features based on the electricity consumption status data; obtains historical similar time periods based on the comprehensive features; inputs the historical similar time periods into a pre-trained prediction model for parameter adjustment; the adjusted prediction model outputs predicted power service data for the test time period; and adjusts the power dispatching scheme based on the predicted power service data, which includes power load data, power price data, and generator output power data. This allows for more accurate prediction of power service data, thus better serving the power dispatching scheme and achieving better adjustment results. Attached Figure Description

[0041] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0042] Figure 1 This paper illustrates a flowchart of a power business data prediction method according to an embodiment of this application.

[0043] Figure 2 This paper illustrates a schematic diagram of a two-stage similar time period calculation process according to an embodiment of this application.

[0044] Figure 3This illustration shows a clustering result diagram according to an embodiment of this application;

[0045] Figure 4 A schematic diagram of the structure of a power business data prediction device according to an embodiment of this application is shown. Detailed Implementation

[0046] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0047] The components of the embodiments of this application described and illustrated in the accompanying drawings can be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of this application provided in the drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0048] In the following text, the terms "comprising," "having," and their cognates, which may be used in various embodiments of this application, are intended only to indicate a particular feature, number, step, operation, element, component, or combination thereof, and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations thereof, or adding the possibility of one or more combinations thereof. Furthermore, the terms "first," "second," "third," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.

[0049] Unless otherwise specified, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which the various embodiments of this application pertain. Terms (such as those defined in commonly used dictionaries) shall be interpreted as having the same meaning as in their contextual meaning in the relevant technical field and shall not be construed as having an idealized or overly formal meaning, unless clearly defined in the various embodiments of this application.

[0050] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0051] To improve the prediction results of power business data and thus enable precise adjustment of power dispatching schemes, this application provides a power business data prediction method, comprising: acquiring electricity consumption status data for a time period to be measured; constructing comprehensive features based on the electricity consumption status data; acquiring historical similar time periods for the time period to be measured based on the comprehensive features; inputting the historical similar time periods into a pre-trained prediction model for parameter adjustment; and outputting predicted power business data for the time period to be measured by the adjusted prediction model; and adjusting the power dispatching scheme based on the predicted power business data, wherein the power business data includes power load data, power price data, and generator output power data.

[0052] The following examples illustrate this method for predicting power business data.

[0053] Figure 1 A flowchart of a power business data prediction method according to an embodiment of this application is shown. Exemplarily, the power business data prediction method includes the following steps:

[0054] Step S100: Obtain power consumption status data for the time period to be tested.

[0055] The technical solution of this embodiment is used to predict electricity business data for a future period of time. Exemplarily, this electricity business data can include electricity prices, electricity consumption, etc. Taking electricity price prediction as an example, electricity price prediction is a crucial link in energy economics and the power system, and has significant guiding significance for electricity trading and power allocation. Electricity prices are refreshed at predetermined intervals, such as 15 minutes. Therefore, when predicting the electricity price for a future day, what is obtained is a price change curve for that day formed at 15-minute intervals.

[0056] To better predict electricity business data, electricity consumption data for the target time period will be obtained. To obtain the electricity price from 8:00 AM to 12:00 AM tomorrow, multiple target time periods will be obtained at preset intervals (e.g., 15 minutes). Electricity consumption data for these target time periods will also be obtained simultaneously.

[0057] It is understandable that future electricity consumption data must be predicted in some way. The electricity consumption data mentioned in this embodiment includes meteorological data, load data, power generation data, and power flow restriction data. Meteorological data can be obtained and predicted through meteorological websites, while load data, power generation data, and power flow restriction data can be calculated and predicted using predetermined power generation and consumption plans. It is understood that different weather conditions, different electricity loads, different power generation amounts, and different power flow restriction data will all affect electricity consumption relationships, thereby affecting electricity prices. These influencing factors all need to be collected and predicted.

[0058] Step S200: Based on the electricity consumption status data, construct comprehensive features, and based on the comprehensive features, obtain historical similar time periods for the time period to be measured.

[0059] like Figure 2 As shown, this step includes the following steps:

[0060] Step S210: According to a preset time interval, acquire multiple historical time periods, as well as historical electricity consumption status data and historical power business data corresponding to the multiple historical time periods.

[0061] In addition, although electricity prices are affected by the above factors, they are also related to past history. Therefore, this embodiment also uses a similar day model to predict future electricity prices. To this end, after determining the future time period to be measured and a large amount of electricity consumption data, it is also necessary to extract the past historical time periods and obtain the time periods similar to the time period to be measured from the historical time periods. For this purpose, it is necessary to use historical similar time periods based on electricity consumption data and the time period to be measured.

[0062] Specifically, you can first obtain a period of time in the past, such as the past month, and then find a similar period of time to the test time from the past month.

[0063] First, based on a preset time interval (e.g., 15 minutes), multiple historical time periods are acquired, along with corresponding historical electricity consumption status data and historical power service data. The time interval here is the same as the time interval used when acquiring the time period to be measured.

[0064] Step S220: Obtain historical characteristics based on the historical electricity consumption status data and historical power business data.

[0065] Based on the historical features and the comprehensive features, obtain multiple historical similar time periods with the highest similarity to the time period to be tested.

[0066] First, let me explain how the above-mentioned comprehensive features are constructed. The comprehensive features and historical features here are features with the same construction components and the same construction method, which will be explained uniformly here.

[0067] Based on the time period to be measured, and the peak and valley characteristics of electricity prices during that time period, time features can be obtained. Similarly, for other data, such as load data, generation data, and power flow constraint data, corresponding feature data can be obtained respectively.

[0068] By comparing historical features and comprehensive features, we can obtain the feature similarity of corresponding features.

[0069] As an example, the temporal similarity is constructed as follows:

[0070]

[0071]

[0072] in Let be the time similarity between the i-th historical time period and the time period to be tested. The date factor represents the historical time period; the closer the historical time period is to the time period being measured, the larger the date factor. This is the attenuation factor for each additional day of distance between a historical date and the date of the test period. This represents the number of time periods from the i-th historical time period to the time period to be tested, and int() is the integer function; Representing the peak-valley factor, if the historical time period and the time period to be measured are at the same peak, flat, or trough, the larger the peak-valley factor, The attenuation coefficient is used to determine whether the historical period and the predicted period are in the same peak, flat period or trough. The holiday factor represents the historical period and the period to be tested falling on the same holiday. The larger the holiday factor, the better. This is the attenuation coefficient for whether the historical period and the predicted period fall within the same holiday or non-holiday period.

[0073] The feature similarity is constructed as follows:

[0074] If the vector of the i-th historical time period is The vector of the time period to be measured is Then we have:

[0075]

[0076] in The feature similarity between the i-th historical time period and the time period to be tested. It is an improved cosine similarity used to measure the similarity between two vectors.

[0077] The grid congestion similarity is constructed as follows:

[0078] The grid congestion similarity is calculated by fusing the effects of line congestion status and node power imbalance, while also considering time trend consistency, using the following formula:

[0079]

[0080] In the formula: The value is the grid congestion similarity, ranging from [0,1]. The closer the value is to 1, the more similar the congestion status is between the historical time period and the time period to be tested. It is a comprehensive index of power grid congestion status at time t during a historical time period (denoted as H); The comprehensive index of power grid congestion status at time t during the time period to be measured (denoted as P); Let be the weighting coefficients at time t, satisfying The peak time value is higher than the valley time value, and is determined based on the experience of power grid dispatchers; T is the total number of times included in the time period; The rate of change of the comprehensive index of congestion status at time t in the historical time period. ; The percentage change rate of the comprehensive index of the congestion status at time t during the test period is given. ; Within the historical period The maximum value; Within the time period to be tested The maximum value.

[0081] Congestion Status Comprehensive Index Taking into account both line congestion and node power imbalance, the formula is as follows:

[0082]

[0083] In the formula: The weighting coefficient for line congestion satisfies The number of transmission lines in the power grid is determined based on the experience of power grid dispatchers; L is the total number of transmission lines in the power grid. Let be the congestion index of the l-th line at time t; N is the total number of nodes in the power grid; The power imbalance contribution of the nth node at time t is dimensionless.

[0084] Line congestion index The formula for quantifying the congestion risk and power flow stability of a single line is as follows:

[0085]

[0086] In the formula: Let l be the importance weight coefficient of the l-th line, satisfying Higher values ​​are taken for tie lines and load center power supply lines, determined based on the experience of power grid dispatchers; This is a weighting coefficient for the utilization rate of the trend. The weighting coefficients for tidal fluctuations satisfy... It is determined based on the experience of power grid dispatchers; This represents the actual active power flow of the l-th line at time t; This is the rated transmission capacity of the l-th line; Let be the absolute value of the actual power flow of the l-th line at time t;

[0087] Let be the power flow utilization rate of the l-th line at time t.

[0088] Let be the power flow utilization rate of the l-th line at time t-1.

[0089] Contribution of node power imbalance The indirect impact of quantified nodes on the overall network power flow distribution and congestion is expressed by the following formula:

[0090]

[0091] In the formula: Let n be the importance weight coefficient of the nth node, satisfying It is determined based on the experience of power grid dispatchers; Let be the net injection power of the nth node at time t. = Node power supply - Node load power (positive indicates power flowing out of the node); Let be the absolute value of the net injected power at the nth node at time t; It represents the sum of the absolute values ​​of the net injected power of all nodes in the entire network at time t.

[0092] The dual-modal adaptive comprehensive similarity model is constructed as follows:

[0093]

[0094] ;

[0095] The objective function aims to minimize the deviation between the overall similarity and the order of temporal similarity, feature similarity, and blocking similarity, and the numerical value. For the i-th comprehensive similarity, Indicates in The position of the l-th element in the vector within the sorted sequence. Indicates in The position of the l-th element in the vector within the sorted sequence. and They represent in Vector sum The vector represents the position of the l-th element in the sorting, and N is the total number of historical time periods. The model optimization objective considers both the order relationship and the numerical value of similarity, and integrates the similarity features of each time period.

[0096] After determining the overall similarity, the historical time periods can be sorted from high to low based on their overall similarity, and then clustered using a clustering algorithm. All samples within the cluster with the highest overall similarity are considered the initial similarity segments, i.e., the first similar time periods.

[0097] Step S230: Based on the historical features and the comprehensive features, obtain multiple historical similar time periods with the highest similarity to the time period to be tested.

[0098] Exemplary, such as Figure 3 The diagram shown is a schematic of the results of a k-clustering algorithm. The points marked with the same color are samples that are clustered into one class. In this embodiment, the time period is the same as the cluster centroid. In this embodiment, the class with the highest overall similarity can be regarded as the sample closest to the cluster centroid as the first similar time period.

[0099] It is understandable that the aforementioned comprehensive similarity has a temporal sequence, therefore Figure 3 The horizontal axis feature 1 and the vertical axis feature 2 can represent the comprehensive similarity and the time series feature corresponding to the comprehensive similarity, respectively.

[0100] After obtaining the first similar time period, this embodiment will also perform a second screening. During the second screening, based on the multi-dimensional temporal features of the first similar time period and the time period to be tested, the key similar features are focused through an attention mechanism, and the finely screened time period is automatically obtained by combining mutual information quantization nonlinear correlation and kernel density estimation. The specific steps are as follows:

[0101] For the first similar time period, temporal features directly related to the aforementioned multidimensional similarity discrimination are extracted to construct a temporal feature matrix, ensuring that the screening logic is consistent with the previous similarity discrimination dimensions: each initial screening time period corresponds to one temporal feature vector, and the vector dimensions cover three types of time series sequences. Time-related time series are determined based on peak and valley state time series and holiday state time series based on "time similarity"; feature-related time series are determined based on meteorological time series, load time series, and power generation time series based on "feature similarity"; congestion-related time series are determined based on congestion state time series based on "congestion similarity".

[0102] Normalization is performed on all the time-series features mentioned above to eliminate the interference of dimensional differences on similarity calculation.

[0103] By automatically learning the weights of each dimension in the temporal features through an attention mechanism, the contribution of key features to similarity discrimination is enhanced.

[0104] Based on the attention weights that enhance the similarity contribution of key features, the mutual information value between the first similar time period and the time period to be tested is calculated to quantify the nonlinear association.

[0105] Kernel density estimation is used to model the distribution of mutual information values ​​for all first similar time periods, automatically locating core similar samples and obtaining refined screening time periods. The mutual information values ​​are used as input, and kernel density estimation is employed to construct a probability density function for the mutual information values. The kernel function type of the kernel density estimation is automatically adapted based on the distribution characteristics of the mutual information values, and the bandwidth is automatically calculated based on the mutual information value distribution using data statistical rules. The mutual information value interval corresponding to the peak value of the probability density function is identified. The mutual information values ​​of the initial screening time periods within this interval are concentrated and highest, representing the optimal correlation with the time period to be tested. All first similar time periods within the mutual information value interval are selected as the final refined screening similar time periods, and the screening boundary is automatically determined by the half-width at half-maximum (WHM) of the probability density function.

[0106] The final refined similarity time period returned at this point is the second similarity time period obtained after the second screening. It can be seen that the second similarity time period obtained through the above method is a time period more relevant to the time period to be tested. This second similarity time period is the historical similarity time period.

[0107] S300, the historical similar time period is input into the pre-trained prediction model for parameter adjustment, and the prediction model after parameter adjustment outputs the predicted power business data for the time period to be tested.

[0108] After obtaining the second similar time period, it is input into a pre-trained prediction model (which could be a Bagging-XGBoost model). This model is then trained and its parameters tuned using the second similar time period. The tuned model is then used for prediction. In this embodiment, the above process is repeated each time a prediction is needed for a future time period. Through this secondary filtering step, historical similar time periods are obtained, and the parameters of the prediction model are then improved. After improvement, the model can predict electricity prices for the target time period based on the data from the second similar time period.

[0109] This embodiment adaptively calculates the comprehensive similarity of historical time periods, uses a clustering algorithm to obtain the first-stage similar time periods, and then uses the mutual information-kernel density estimation method to obtain the first-order similar time periods that meet the matching conditions as the second-stage similar time periods. By considering the introduction of time similarity based on whether they are the same peak and valley, electricity consumption characteristic similarity, and congestion similarity constructed based on power flow and limit data, and by constructing a bimodal adaptive comprehensive similarity model that considers the similarity value and order relationship, the accuracy of similarity calculation is significantly improved, effectively taking into account the subjective cognition of decision-makers. Finally, based on the initial screening, the mutual information-kernel density estimation method is further applied to achieve a second fine screening, thereby greatly improving the accuracy of similar time period selection and enhancing the accuracy of electricity price forecasting.

[0110] Step S400: Adjust the power dispatching scheme based on the predicted power business data, which includes power load data, power price data, and generator output power data.

[0111] It is understood that power business data includes power load data, power price data, and generator output power data. All of the above-mentioned power business data can be included in the prediction scope of this scheme. Predicting different power business data can adjust and improve different power dispatch schemes.

[0112] As an example, changes in electricity business data influence power dispatching schemes. The more accurate the electricity price forecast, the better the adjustment support for the power dispatching scheme. There are various types of power dispatching schemes. For any scheme, the power generation, power consumption, power distribution, generator unit combination, peak shaving, frequency regulation, and reserve plan arrangements for the time period to be measured can be adjusted based on the range of the predicted electricity business data.

[0113] For example, optimizing unit combination and output allocation requires advance planning of start-up, shutdown, and output plans for different types of units (thermal power, hydropower, gas-fired power, renewable energy, etc.) (i.e., "unit combination"). Electricity price is the core basis for judging the "match between generation costs and revenue": Day-ahead dispatch (1 day in advance): For units with "high start-up and shutdown costs and slow adjustment," such as thermal power and nuclear power, start-up and shutdown plans need to be formulated based on day-ahead electricity price forecasts. For example, if the predicted morning peak electricity price is higher than the marginal cost (fuel + operation and maintenance costs) of thermal power units, dispatch will prioritize full-capacity operation of thermal power units; if the predicted peak output of renewable energy at midday is lower (supply exceeds demand), the output of thermal power units will be reduced to avoid electricity waste.

[0114] For example, it can improve resource allocation efficiency by accurately predicting the overlap between peak photovoltaic power output and off-peak residential load. Dispatch can then guide energy storage to charge at full capacity, avoiding the abandonment of new energy sources, while thermal power plants maintain optimal load operation, reducing energy consumption and generating economic benefits. Another example is industrial users who, based on "95% accurate electricity price forecasts," determine that "2-6 AM is the low-price period," allowing them to confidently start all production lines. This reduces their own electricity costs and helps the grid absorb off-peak renewable energy output, achieving a win-win situation for "users, grid, and renewable energy"—the core objective of coordinated dispatch of power generation, grid, load, and storage.

[0115] Figure 4 A schematic diagram of a power business data forecasting apparatus according to an embodiment of this application is shown. Exemplarily, the apparatus includes:

[0116] The data acquisition module 10 is used to acquire power consumption status data for the time period to be tested;

[0117] The feature construction module 20 is used to construct comprehensive features based on the electricity consumption status data, and to obtain historical similar time periods of the time period to be measured based on the comprehensive features.

[0118] The model adjustment module 30 is used to input the historical similar time period into the pre-trained prediction model for parameter adjustment, and the prediction model after parameter adjustment outputs the predicted power business data of the time period to be tested.

[0119] The scheduling and adjustment module 40 is used to adjust the power dispatching scheme based on the predicted power business data, which includes power load data, power price data, and generator output power data.

[0120] It is understood that the apparatus of this embodiment corresponds to the method of the above embodiments, and the options in the above embodiments are also applicable to this embodiment, so they will not be described again here.

[0121] The processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including at least one of a Central Processing Unit (CPU), Graphics Processing Unit (GPU), Network Processor (NP), Digital Signal Processor (DSP), Application-Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application.

[0122] The memory can be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The memory is used to store computer programs, and the processor can execute the computer programs accordingly after receiving execution instructions.

[0123] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that, in alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0124] In addition, the functional modules or units in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0125] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a smartphone, personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0126] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A power business data prediction method characterized by, The method comprises the following steps: acquiring power consumption state data of a to-be-tested time period; constructing comprehensive features according to the power consumption state data, and acquiring historical similar time periods of the to-be-tested time period according to the comprehensive features; inputting the historical similar time periods into a pre-trained prediction model for parameter adjustment, and then outputting predicted power service data of the to-be-tested time period by using the adjusted prediction model; adjusting a power dispatching scheme according to the predicted power service data, wherein the power service data comprises power load data, power price data, and generator set output power data; the step of acquiring historical similar time periods of the to-be-tested time period according to the comprehensive features comprises the following steps: acquiring a plurality of historical time periods, and historical power consumption state data and historical power service data corresponding to the plurality of historical time periods according to a preset time interval; obtaining historical features according to the historical power consumption state data and the historical power service data; acquiring a plurality of historical similar time periods with the highest similarity to the to-be-tested time period based on the historical features and the comprehensive features; the step of acquiring a plurality of historical similar time periods with the highest similarity to the to-be-tested time period based on the historical features and the comprehensive features comprises the following steps: obtaining comprehensive similarity of each historical time period relative to the to-be-tested time period based on the comprehensive features and each historical feature; performing clustering on each comprehensive feature by using a mean clustering algorithm to obtain a plurality of historical time period sets, and each historical time period in a historical time period set with the highest comprehensive similarity is a first similar time period; performing two-stage screening on a plurality of the first similar time periods by using an attention mechanism enhanced mutual information-kernel density estimation method to obtain a plurality of second similar time periods; the step of performing two-stage screening on a plurality of the first similar time periods by using an attention mechanism enhanced mutual information-kernel density estimation method to obtain a plurality of second similar time periods comprises the following steps: acquiring corresponding time, power consumption feature and congestion time sequence feature vectors according to the first similar time periods; performing normalization operation on all the time sequence feature vectors to obtain dimensionless features; determining the weight of each dimension in the dimensionless features by using an attention mechanism, and calculating the nonlinear relationship between the first similar time and the to-be-tested time period; calculating the mutual information values of all the first similar time periods by using kernel density estimation, constructing the probability density function of the mutual information values, and identifying the mutual information value interval corresponding to the peak value of the probability density function; the first similar time period in the mutual information value interval and exceeding the half-height width of the probability density function is a second similar time period.

2. The electric power business data forecasting method according to claim 1, characterized by, the power consumption state data comprises meteorological data, load data, power generation data and power flow restriction data; the step of constructing comprehensive features according to the power consumption state data comprises the following steps: calculating power consumption feature similarity based on the meteorological data, the load data and the power generation data; obtaining time similarity according to the to-be-tested time period; calculating congestion similarity based on the power flow restriction data; The blockage similarity, time similarity, and electricity consumption feature similarity are sorted to obtain the corresponding order relationship. The comprehensive feature is calculated by combining the blockage similarity, time similarity, electricity consumption feature similarity, and the corresponding order relationship.

3. The electric power business data prediction method according to claim 1, characterized by, The prediction model is one of the following: time series model, neural network model, distributed gradient and ensemble machine learning model, Bagging-XGBoost model.

4. The electric power business data prediction method according to claim 1, characterized by, The adjustment of the power dispatching scheme based on the predicted power business data, wherein the power business data includes power load data, power price data, and generator output power data, including: Based on the range of the predicted power business data, the power generation, power consumption, power distribution, generator set combination, peak shaving, frequency regulation, and reserve plan arrangements for the time period to be tested are adjusted to adjust the power dispatching scheme.

5. An electric power service data prediction device characterized by comprising: include: The data acquisition module is used to acquire electricity consumption status data for the time period to be tested; The feature construction module is used to construct comprehensive features based on the electricity consumption status data, and to obtain historical similar time periods for the time period to be measured based on the comprehensive features. The model adjustment module is used to input the historical similar time period into the pre-trained prediction model for parameter adjustment, and the prediction model after parameter adjustment outputs the predicted power business data for the time period to be tested. The scheduling and adjustment module is used to adjust the power dispatching scheme based on the predicted power business data, which includes power load data, power price data, and generator output power data. The step of obtaining historical similar time periods for the time period to be tested based on the comprehensive features includes: According to a preset time interval, multiple historical time periods are obtained, along with historical electricity consumption status data and historical power business data corresponding to the multiple historical time periods; Historical characteristics are obtained based on the historical electricity consumption status data and historical power business data; Based on the historical features and the comprehensive features, obtain multiple historical similar time periods that have the highest similarity to the time period to be tested; The process of obtaining multiple historically similar time periods with the highest similarity to the time period to be tested, based on the historical features and the comprehensive features, includes: Based on the comprehensive features and each of the historical features, the comprehensive similarity of each historical period relative to the time period to be tested is obtained; The combined features are clustered using the mean clustering algorithm to obtain multiple sets of historical time periods. The historical time periods in the set of historical time periods with the highest combined similarity are the first similar time periods. Multiple first similar time periods are filtered in two stages using an attention-enhanced mutual information-kernel density estimation method to obtain multiple second similar time periods; The mutual information-kernel density estimation method enhanced by the attention mechanism performs a two-stage screening of multiple first similar time periods to obtain multiple second similar time periods, including: Based on the first similar time period, obtain the corresponding time, electricity consumption characteristics and congestion time sequence feature vectors; Normalize all the aforementioned time-series feature vectors to obtain dimensionless features; The weight of each dimension in the dimensionless feature is determined through an attention mechanism, and a nonlinear relationship between the first similar time and the to-be-measured time period is calculated; The mutual information values of all the first similar time periods are calculated through kernel density estimation, and a probability density function of the mutual information values is constructed to identify an interval of mutual information values corresponding to a peak value of the probability density function; First similar time periods in the interval of mutual information values and exceeding a half-height width of the probability density function are second similar time periods.

6. A terminal device, characterized by comprising: The terminal device comprises a processor and a memory, the memory stores a computer program, and the processor is configured to execute the computer program to implement the power business data prediction method in any one of claims 1-4.

7. A computer-readable storage medium, characterized in that, The computer program is stored in the memory and is executed on the processor to implement the power business data prediction method in any one of claims 1-4.

Images